ABSTRACT

We will discuss recent studies demonstrating that computational analysis of transcribed spoken and written language can provide for highly accurate diagnostics over a wide variety of psychiatric and neurologic conditions such as psychosis, drug abuse, Parkinson’s and Alzheimer’s, and in particular surpass the clinical assessment of conversion to psychosis in at-risk patients. These results are based on the mathematical formalization of psychiatric qualitative knowledge related to the characterization of the conditions (e.g. ‘flight of ideas’ in psychosis) and drug effects (e.g. increased intimacy with ecstasy), as well as novel linguistic feature extraction approaches. Moreover, we show novel results using publicly available text sources suggesting that: it is possible to define an embedding space to map and compare different conditions, and computational studies of a public personalities can yield novel insights into normal aging and neurodegenerative disorders. Finally, we discuss the implications of these findings for mental health, and some specific efforts to bring this technology closer to the clinic.

Bio

Guillermo Cecchi received an education in Physics (MSc, University of La Plata, Argentina, 1991), Physics and Biology (PhD, The Rockefeller University, 1994-1999), and Imaging in Psychiatry (Postdoctoral Fellow, Cornell University 2000-2001). He has been interested in diverse aspects of theoretical biology, including Brownian transport, molecular computation, spike reliability in neurons, song production and representation in songbirds, statistics of natural images and visual perception, statistics of natural language, and brain imaging. In 2001 he joined IBM Research to work on computational approaches to brain function.

In recent years, Dr. Cecchi has pioneered the use of a computational linguistics approach to quantify psychiatric conditions from short speech samples, applying it successfully to conditions as diverse as schizophrenia, mania, prodromal psychosis, and drug and alcohol intake.